A spectroscopic approach based on Non-negative Matrix Factorization method for multispectral fluorescence images unmixing

Immunofluorescence is the visualization of organs using antibodies as fluorescent probes. It allows to detect and to localize one or more fluorescent labeled structure using specific antibodies. A major problem in this regard is the autofluorescence present in some biological samples. The spatial localization analysis of interest structures is often perturbed by a significant overlap between the marker fluorescence and the biological tissues autofluorescence. This overlap leads to the formation of a complex multicolored image. Indeed, the surface covered by one pixel of such image may contain more than one fluorescent sources component. Thus, the spectrum observed in each pixel may be a linear mixture of contributions from several sources. A spectroscopic approach, based on the Non-negative Matrix Factorization (NMF) method, is explored to decompose each mixed pixel spectrum into several pure fluorescent spectra and their relative contributions. This blind source separation method treats specific fluorescence and autofluorescence as different sources to separate. Experimental results on real mixtures confirm the effectiveness of the proposed method.